SEQUENTIAL ANALYSIS WITH VARIABLE DECISION BOUNDARIES

THE METHODOLOGY DEVELOPED IS A MODIFICATION OF THE CLASSICAL WALD SEQUENTIAL ANALYSIS TO BE APPLIED WHEN THE INCOMING SAMPLE IS "STRATIFIED," I.E., THERE IS SOME KNOWN NON-RANDOMNESS PRESENT WHICH CAN BE CIRCUMVENTED BY UTILIZATION OF RELEVANT CONCOMITANT INFORMATION. THE CLASSICAL AND MODIFIED TECHNIQUES WERE APPLIED TO 200 SEQUENCES OF STRATIFIED SAMPLES GENERATED BY DIGITAL SIMULATION WITH REGARD TO A NULL AND AN ALTERNATIVE HYPOTHESIS UNDER 9 INCOMING SAMPLE SITUATIONS. DISTRIBUTION OF SAMPLE SIZE, AVERAGE SAMPLE SIZE, VARIANCE OF SAMPLE SIZE, NUMBER OF CORRECT AND INCORRECT DECISIONS WERE RECORDED FOR EACH TECHNIQUE AND ARE COMPARED. A TECHNIQUE IS PRESENTED WHICH MAKES IT POSSIBLE TO PERFORM SEQUENTIAL ANALYSIS ALSO UNDER CONDITIONS SO ADVERSE THAT THE CLASSICAL TECHNIQUE IS APT TO FAIL.